Bridge crack dynamic monitoring system and method based on diffusion probability model

Abstract

The invention relates to the technical field of bridge material analysis, and discloses a bridge crack dynamic monitoring system and method based on a diffusion probability model, and the method comprises the steps: constructing a reflection amplitude matrix and a crack depth matrix according to ultrasonic data; a contrast value, a homogeneity value and a correlation value in a bridge crack image are obtained according to a gray-level co-occurrence algorithm, a texture feature matrix is constructed based on the contrast value, the homogeneity value and the correlation value, a reflection amplitude matrix, a crack depth matrix and the texture feature matrix are constructed respectively, the dynamic change characteristics of the bridge crack are comprehensively represented, and the accuracy of the dynamic change of the bridge crack is improved. The problem of insufficient accuracy and reliability of crack feature extraction in the prior art is solved, the crack propagation rate can be captured in real time by screening the feature matrix sequence based on the W comprehensive feature matrixes and inputting the pre-trained diffusion probability model, and the dynamic change of the crack is monitored.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge material analysis, and more specifically, to a bridge crack dynamic monitoring system and method based on a diffusion probability model. Background Art

[0002] Bridges are important transportation infrastructure, and cracks are one of the common defects in bridge structures. Early detection and dynamic monitoring are crucial to preventing structural failure and extending service life. Traditional bridge crack detection methods mainly include manual visual inspection, acoustic emission technology, ultrasonic detection, and infrared thermal imaging. However, in recent years, with the rapid development of computer vision and machine learning technology, bridge crack detection methods based on image analysis have gradually attracted widespread attention. Although related methods for bridge crack detection using computer vision have been disclosed in the prior art, there are still some problems that need to be solved.

[0003] For example, the Chinese patent application with publication number CN115078382A provides a bridge crack monitoring system based on video images, which consists of a front-end system and a back-end system. The front-end system collects crack image information and sends it to the back-end system via wireless transmission. The back-end system stores and displays the crack image information and supports cloud display and precise calculation. The Chinese patent with authorization announcement number CN118730914B provides a crack monitoring device for a cross-sea bridge and a monitoring method thereof. The system includes an image acquisition unit and a central control unit. The image acquisition unit is used to collect visual image information of bridge cracks. The central control unit receives crack change parameters and image information, performs processing and calculation, and outputs monitoring results.

[0004] Although the prior art has disclosed relevant content on bridge crack detection using computer vision, it is mainly limited to static monitoring of bridge cracks and cannot monitor dynamic changes of cracks. In addition, the texture features of different bridge structures vary greatly, resulting in insufficient accuracy and reliability of crack feature extraction. These problems make it difficult for bridge crack detection to meet actual needs in terms of real-time and dynamic performance, and also affect the accuracy of monitoring results and the effectiveness of bridge structure safety assessment.

[0005] In view of this, the present invention proposes a bridge crack dynamic monitoring system and method based on a diffusion probability model to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a bridge crack dynamic monitoring system and method based on a diffusion probability model.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In a first aspect, a bridge crack dynamic monitoring method based on a diffusion probability model is provided, comprising:

[0009] Acquire ultrasonic data and bridge crack images, construct reflection amplitude matrix and crack depth matrix according to the ultrasonic data, and obtain contrast value, homogeneity value and correlation value in the bridge crack image according to the grayscale symbiosis algorithm;

[0010] A texture feature matrix is ​​constructed based on contrast values, homogeneity values ​​and correlation values, W comprehensive feature matrices are constructed based on the reflection amplitude matrix, the crack depth matrix and the texture feature matrix, and a feature matrix sequence is determined based on the W comprehensive feature matrices;

[0011] The characteristic matrix sequence is input into a pre-trained diffusion probability model to obtain a crack monitoring result output by the diffusion probability model, wherein the crack monitoring result at least includes a crack propagation rate.

[0012] Furthermore, the method for obtaining ultrasonic data includes:

[0013] The bridge crack images are classified to obtain the corresponding crack classification results. If the crack classification result is a longitudinal crack, the ultrasonic probe is arranged along the length direction of the bridge. If the crack classification result is a transverse crack, the ultrasonic probe is arranged perpendicular to the length direction of the bridge. If the crack classification result is an oblique crack, the ultrasonic probe is arranged along the crack direction of the bridge.

[0014] Furthermore, the ultrasonic data includes the peak occurrence time, and the method of constructing the crack depth matrix according to the ultrasonic data includes:

[0015] The crack depth value is determined according to the peak occurrence time in the ultrasonic data and the preset propagation speed, and the crack depth matrix is ​​constructed according to the crack depth value.

[0016] Furthermore, the method for obtaining the contrast value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0017] The contrast value in the direction θ and the co-occurrence probability in the direction θ in the bridge crack image are obtained according to the grayscale co-occurrence algorithm, and the corresponding contrast value is calculated according to the contrast value in the direction θ and the co-occurrence probability in the direction θ.

[0018] Furthermore, the method for obtaining the correlation value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0019] According to the grayscale symbiosis algorithm, the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j in the bridge crack image are obtained, and the corresponding correlation value is calculated according to the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j.

[0020] Furthermore, the method of constructing W comprehensive feature matrices according to the reflection amplitude matrix, the crack depth matrix and the texture feature matrix includes:

[0021] Adjust the value of K, obtain W texture feature matrices, traverse the W texture feature matrices, align the reflection amplitude matrix, crack depth matrix and texture feature matrix by mathematical transformation, perform weighted fusion on the reflection amplitude matrix, crack depth matrix and texture feature matrix, and construct W comprehensive feature matrices. K represents the value angle corresponding to the texture feature matrix.

[0022] Furthermore, the method for determining the feature matrix sequence based on W comprehensive feature matrices includes:

[0023] The crack direction range in the bridge crack image is determined according to the Hough transform algorithm, and P comprehensive feature matrices are screened out from W comprehensive feature matrices according to the crack direction range, and the P comprehensive feature matrices are used as a feature matrix sequence. The crack direction range is the range of the crack extension direction in the bridge crack image.

[0024] Furthermore, the method of selecting P comprehensive feature matrices from W comprehensive feature matrices according to the crack direction range includes:

[0025] The value range corresponding to each comprehensive feature matrix is ​​obtained to determine whether the value range belongs to the crack direction range. If not, the corresponding comprehensive feature matrix is ​​eliminated to screen out P comprehensive feature matrices, and the value range is the value range of K.

[0026] Furthermore, the training method of the diffusion probability model includes:

[0027] Pair the historical feature matrix sequence with the corresponding historical crack monitoring result label to construct a training set, use the historical feature matrix sequence as the input of the sub-model, and use the corresponding historical crack monitoring result as the output of the sub-model;

[0028] The total number of diffusion steps is preset to T, and the noise increment of each step is defined as β t For each data point in the training set, the current diffusion step t and the corresponding noise data x are generated by random sampling. t , according to the current diffusion step t and the noise data x t Construct the loss function, the noise data x t The distribution form of is Gaussian distribution;

[0029] Randomly extract a batch of data points from the training set and generate corresponding noise data x for each data point t, calculate the corresponding loss function value, update the sub-model parameters through the back-propagation algorithm and minimize the loss function value as the goal, and iteratively train the sub-model to obtain the diffusion probability model.

[0030] In a second aspect, a bridge crack dynamic monitoring system based on a diffusion probability model is provided, which is used to implement the above-mentioned bridge crack dynamic monitoring method based on a diffusion probability model, including:

[0031] Data processing module: used to obtain ultrasonic data and bridge crack images, construct reflection amplitude matrix and crack depth matrix according to ultrasonic data, and obtain contrast value, homogeneity value and correlation value in bridge crack images according to grayscale symbiosis algorithm;

[0032] Matrix construction module: used to construct a texture feature matrix based on contrast value, homogeneity value and correlation value, construct W comprehensive feature matrices according to reflection amplitude matrix, crack depth matrix and texture feature matrix, and determine feature matrix sequence based on W comprehensive feature matrices;

[0033] Monitoring module: used to input the feature matrix sequence into the pre-trained diffusion probability model to obtain the crack monitoring result output by the diffusion probability model, and the crack monitoring result at least includes the crack propagation rate.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention combines ultrasonic data and image texture features of bridge crack images to construct reflection amplitude matrix, crack depth matrix and texture feature matrix respectively, which comprehensively characterizes the dynamic change characteristics of bridge cracks and solves the problem of insufficient accuracy and reliability of crack feature extraction in the prior art. By screening feature matrix sequences based on W comprehensive feature matrices and inputting pre-trained diffusion probability models, the crack propagation rate can be captured in real time, and the dynamic changes of cracks can be monitored, which solves the problem of insufficient dynamic monitoring capability in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a bridge crack dynamic monitoring method based on a diffusion probability model in the present invention;

[0037] Figure 2 It is a structural schematic diagram of a bridge crack dynamic monitoring system based on a diffusion probability model in the present invention;

[0038] Figure 3 A schematic diagram of a flow chart of a method for acquiring ultrasonic data in the present invention;

[0039] Figure 4Schematic diagram of the flow of the method for determining a feature matrix sequence based on W comprehensive feature matrices in the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Example 1

[0042] See also Figure 1 As shown, this embodiment discloses a bridge crack dynamic monitoring method based on a diffusion probability model, including:

[0043] S10: Acquire ultrasonic data and a bridge crack image, construct a reflection amplitude matrix and a crack depth matrix according to the ultrasonic data, and acquire a contrast value, a homogeneity value, and a correlation value in the bridge crack image according to a grayscale symbiosis algorithm;

[0044] In this embodiment, in the process of acquiring ultrasonic data and bridge crack images, the bridge to be monitored can be in a state of being open to traffic or in a state of being closed to traffic. The bridge crack image can be acquired by a camera installed on the surface of the bridge. The ultrasonic data can be acquired by arranging corresponding ultrasonic probes in a fixed manner. The fixed manner means that the ultrasonic probes are arranged in key parts of the bridge structure according to preset rules and positions, such as crack-prone areas or structural nodes with greater stress. It can also be acquired in the following manner:

[0045] like Figure 3 As shown, the method for obtaining ultrasonic data includes:

[0046] The bridge crack images are classified to obtain the corresponding crack classification results. If the crack classification result is a longitudinal crack, the ultrasonic probe is arranged along the length direction of the bridge. If the crack classification result is a transverse crack, the ultrasonic probe is arranged perpendicular to the length direction of the bridge. If the crack classification result is an oblique crack, the ultrasonic probe is arranged along the crack direction of the bridge.

[0047] It should be added that the bridge crack images can be classified through existing classification models, such as convolutional neural networks (CNN) or support vector machines (SVM). These models can be classified based on the texture, shape, and direction of the crack images. In addition, pre-trained deep learning models (such as ResNet or VGG) can be used to extract and classify the features of the bridge crack images to quickly obtain the specific direction and type of the cracks.

[0048] It can be understood that, compared with the fixed arrangement of ultrasonic probes, the advantage of arranging ultrasonic probes in the above manner is that it can ensure that the arrangement of ultrasonic probes is more targeted at the actual direction of the crack, thereby improving the accuracy of the reflection amplitude matrix and the crack depth matrix in subsequent steps. At the same time, this arrangement method can more effectively capture the texture features related to the crack direction, which helps to make the covariance value of the comprehensive feature matrix constructed in subsequent steps more accurate, and ultimately optimize the determination of the feature matrix, thereby improving the accuracy and reliability of the diffusion probability model in crack propagation rate monitoring.

[0049] In this embodiment, the ultrasonic data includes at least the maximum amplitude value and the peak occurrence time. The maximum amplitude value refers to the maximum amplitude of the waveform signal reflected back during the propagation of the ultrasonic signal, and the peak occurrence time refers to the time point corresponding to the maximum amplitude value in the ultrasonic signal.

[0050] Methods for constructing a reflection amplitude matrix from ultrasonic data include:

[0051]

[0052] Among them, U I represents the reflection amplitude matrix, I M,N It is characterized by the maximum amplitude value measured by the transmission and reception of ultrasonic signals between the Mth ultrasonic probe and the Nth ultrasonic probe.

[0053] Methods for constructing a crack depth matrix from ultrasonic data include:

[0054] The crack depth value is determined according to the peak occurrence time in the ultrasonic data and the preset propagation speed, and the crack depth matrix is ​​constructed according to the crack depth value.

[0055] Methods for constructing a crack depth matrix based on crack depth values ​​include:

[0056]

[0057] Among them, U D represents the crack depth matrix, D M,N It is characterized by the crack depth value corresponding to the transmission and reception of ultrasonic signals between the Mth ultrasonic probe and the Nth ultrasonic probe.

[0058] It can be understood that the propagation speed mentioned above refers to the speed at which ultrasound waves propagate in bridge materials, which is usually determined by the physical properties of the material (such as density and elastic modulus). Calculating the crack depth value is a prior art and will not be elaborated in this embodiment.

[0059] The method of obtaining the contrast value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0060] The contrast value in the direction θ and the co-occurrence probability in the direction θ in the bridge crack image are obtained according to the grayscale co-occurrence algorithm, and the corresponding contrast value is calculated according to the contrast value in the direction θ and the co-occurrence probability in the direction θ.

[0061] The method of calculating the corresponding contrast value according to the contrast value in the direction θ and the co-occurrence probability in the direction θ includes:

[0062] Cot θ =∑ i,j (ij) 2 P θ (i,j);

[0063] Among them, Cot θ is the contrast value in direction θ, P θ (i, j) represents the co-occurrence probability of a pixel pair with grayscale value i and grayscale value j in direction θ.

[0064] It should be noted that the grayscale co-occurrence algorithm refers to a method of describing the texture characteristics of an image by statistically analyzing the grayscale value relationship between pixels in the image and constructing a grayscale co-occurrence matrix. It reflects the spatial distribution law and mutual relationship of different grayscale values ​​in the image by calculating the grayscale value combination frequency of pixels in the image at a specific direction and distance. The larger the contrast value, the more drastic the change in the grayscale value in the image and the more obvious the edge features of the cracks. In bridge crack detection, areas with larger contrast values ​​usually correspond to significant parts of the cracks. Therefore, the larger the contrast value, the clearer the characteristics of the crack area can be highlighted, which helps to improve the accuracy of crack detection and the effectiveness of feature extraction.

[0065] The method of obtaining the homogeneity value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0066]

[0067] Among them, Hoy θ is the homogeneity value in the direction θ.

[0068] It should be noted that in this embodiment, the larger the homogeneity value, the more uniform the grayscale distribution in the bridge crack image, and the smoother the texture change in the crack area, which usually reflects that the grayscale difference in the area around the crack is small, making it difficult to highlight the characteristics of the crack area.

[0069] The method for obtaining the correlation value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0070] According to the grayscale symbiosis algorithm, the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j in the bridge crack image are obtained, and the corresponding correlation value is calculated according to the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j.

[0071] The method of calculating the corresponding correlation value according to the mean of the gray value i, the mean of the gray value j, the standard deviation of the gray value i, and the standard deviation of the gray value j includes:

[0072]

[0073] Among them, Con θ is the correlation value in direction θ, μ i Represents the mean value of gray value i, μ j Represents the mean value of gray value j, σ i Characterizes the standard deviation of gray value i, σ j Characterizes the standard deviation of the gray value j.

[0074] It should be noted that the mean of the grayscale value i refers to the expected value of the second grayscale value j when the first grayscale value is i. Similarly, the mean of the grayscale value j refers to the expected value of the second grayscale value i when the first grayscale value is j. The mean of the grayscale value i is μ i It reflects the average grayscale value of the pixels around the pixel with grayscale value i in the image. i This indicates that the pixels with gray value i tend to have higher gray value, while the pixels with lower μ i This indicates a tendency towards lower gray values.

[0075] In this embodiment, σ i and σ j It describes the extent of grayscale value distribution around pixels with grayscale values ​​i and j. A higher standard deviation means a larger range of grayscale value variation and a rougher texture; a lower standard deviation means a smaller grayscale value variation and a finer texture. From the above, it can be seen that in this implementation, the larger the correlation value, the more consistent the variation trend of the grayscale value in the bridge crack image, and the stronger the correlation of the texture in the crack area, which helps to clearly distinguish the overall morphological characteristics of the crack. For subsequent crack monitoring, this can improve the accuracy and reliability of crack feature extraction, thereby more accurately calculating the crack propagation rate and morphological changes, and enhancing the effectiveness and accuracy of the monitoring results.

[0076] S20: constructing a texture feature matrix based on the contrast value, the homogeneity value and the correlation value, constructing W comprehensive feature matrices according to the reflection amplitude matrix, the crack depth matrix and the texture feature matrix, and determining a feature matrix sequence based on the W comprehensive feature matrices;

[0077] Methods for constructing a texture feature matrix based on contrast values, homogeneity values, and correlation values ​​include:

[0078] The contrast values, homogeneity values ​​and correlation values ​​in K directions in the bridge crack image are calculated according to the grayscale symbiosis algorithm, and the contrast values, homogeneity values ​​and correlation values ​​in different directions are combined respectively to construct the texture feature matrix.

[0079] For example, if K=4, the four directions are 0°, 45°, 90° and 135°, respectively, and the texture feature matrix is ​​as follows:

[0080]

[0081] Among them, TFM represents texture feature matrix.

[0082] The method of constructing W comprehensive feature matrices according to the reflection amplitude matrix, the crack depth matrix and the texture feature matrix includes:

[0083] Adjust the value of K, obtain W texture feature matrices, traverse the W texture feature matrices, align the reflection amplitude matrix, crack depth matrix and texture feature matrix by mathematical transformation, perform weighted fusion on the reflection amplitude matrix, crack depth matrix and texture feature matrix, and construct W comprehensive feature matrices. K represents the value angle corresponding to the texture feature matrix.

[0084] From the above content, it can be seen that the value of K can be 0°, 30°, 45°, 60°, etc., so the value of K is adjusted to construct W texture feature matrices. The purpose of the size alignment through mathematical transformation is to ensure that the texture feature matrices in different directions and scales are consistent in dimension and structure, so as to facilitate the subsequent effective fusion with the reflection amplitude matrix and the crack depth matrix. The fusion of the reflection amplitude matrix, the crack depth matrix and the texture feature matrix requires that all matrices can be effectively weighted superimposed or operated in structure. If the matrix dimensions are inconsistent, direct operation will lead to calculation errors. The weighted fusion mentioned above refers to the weighted superposition of the corresponding elements in each matrix. It should be noted that weighted superposition refers to the splicing on the vector. Vector splicing is to connect multiple matrices or vectors according to a certain dimension (such as columns or rows) to form a new matrix.

[0085] Exemplarily, the construction of the comprehensive feature matrix is ​​as follows:

[0086] U cod =w1·U I +w2·U D +w3·TFM;

[0087] Among them, Ucod is a comprehensive feature matrix, w1, w2, and w3 are all weight factors.

[0088] like Figure 4 As shown, the method for determining the feature matrix sequence based on W comprehensive feature matrices includes:

[0089] The crack direction range in the bridge crack image is determined according to the Hough transform algorithm, and P comprehensive feature matrices are screened out from W comprehensive feature matrices according to the crack direction range, and the P comprehensive feature matrices are used as a feature matrix sequence. The crack direction range is the range of the crack extension direction in the bridge crack image.

[0090] It can be understood that since the present invention requires dynamic detection of bridge cracks, the crack direction range in the bridge crack image is determined by the Hough transform algorithm, and P comprehensive feature matrices related to the crack extension direction are selected from the W comprehensive feature matrices as the feature matrix sequence, which can effectively reduce the interference of irrelevant features and improve the accuracy and efficiency of dynamic crack detection. This method ensures that the feature matrix sequence can more accurately reflect the actual directional characteristics of crack extension, thereby providing more accurate and dynamic support for the calculation of crack extension rate, monitoring of morphological changes, and model prediction.

[0091] The method of selecting P comprehensive feature matrices from W comprehensive feature matrices according to the crack direction range includes:

[0092] The value range corresponding to each comprehensive feature matrix is ​​obtained to determine whether the value range belongs to the crack direction range. If not, the corresponding comprehensive feature matrix is ​​eliminated to screen out P comprehensive feature matrices, and the value range is the value range of K.

[0093] It can be understood that the value of K can be 0°, 30°, 45° and 60°, etc., then the crack direction range is also an angle range. For example, the crack direction range is 30° to 90°. If the values ​​of K corresponding to the first comprehensive feature matrix include 30°, 45° and 60°, and the values ​​of K corresponding to the second comprehensive feature matrix include 60°, 90° and 135°, then the second comprehensive feature matrix is ​​eliminated and the first comprehensive feature matrix is ​​retained. The purpose of this is to ensure that the feature matrix sequence selected from the comprehensive feature matrix is ​​highly correlated with the crack direction range, thereby effectively reducing the interference of irrelevant features on the monitoring of crack extension direction, and improving the accuracy and efficiency of dynamic detection of bridge cracks. At the same time, this method can more accurately reflect the actual directional characteristics of crack extension, and provide more accurate and dynamic support for the calculation of crack extension rate, monitoring of morphological changes and model prediction.

[0094] S30: inputting the feature matrix sequence into a pre-trained diffusion probability model to obtain a crack monitoring result output by the diffusion probability model, wherein the crack monitoring result at least includes a crack propagation rate;

[0095] The training methods of the diffusion probability model include:

[0096] Pair the historical feature matrix sequence with the corresponding historical crack monitoring result label to construct a training set. The historical feature matrix sequence is used as the input of the sub-model, and the corresponding historical crack monitoring result is used as the output of the sub-model. The sub-model can be U-Net or Transformer.

[0097] The total number of diffusion steps is preset to T, and the noise increment of each step is defined as β t For each data point in the training set, the current diffusion step t and the corresponding noise data x are generated by random sampling. t , according to the current diffusion step t and the noise data x t Construct the loss function, the noise data x t The distribution form of is Gaussian distribution;

[0098] Randomly extract a batch of data points from the training set and generate corresponding noise data x for each data point t , calculate the corresponding loss function value, update the sub-model parameters through the back-propagation algorithm and minimize the loss function value as the goal, and iteratively train the sub-model to obtain the diffusion probability model.

[0099] It should be noted that in the training of the diffusion probability model, the design of each step is to ensure that the model can efficiently learn the mapping relationship between the crack feature sequence and the crack propagation rate. t This method can gradually perturb the complex information in the feature matrix sequence to simulate the uncertainty and randomness in the real bridge crack data. This process of gradually adding noise can enhance the generalization ability of the model after back propagation and adapt to the crack expansion in different environments.

[0100] Generate noisy data x t The methods include:

[0101]

[0102] Among them, α t is the forward process coefficient, is the cumulative coefficient, x0 is the data point, ∈ is the standard normal distribution noise, I represents the unit matrix, α s is the cumulative coefficient corresponding to step s, Characterized by normal distribution.

[0103] According to the current diffusion step t and the noise data x t Methods for constructing loss functions include:

[0104]

[0105] in, Characterized as a single-step loss, Characterized as the total loss, It is represented as the expected value of the initial data point x0, the standard normal distribution noise ∈ and the current diffusion step t, ||·|| 2 Characterized as the square norm, ∈ θ (x t ,t) represents the predicted normal distribution noise at the tth step.

[0106] It should be noted that the loss function constructed based on the current diffusion step t is obtained by comparing the difference between the output generated by the current model and the noise component in the real data. The process of minimizing the loss function value can help the model gradually restore the true structure of the feature matrix sequence and ensure that the model has an accurate understanding and prediction capability of the dynamic expansion characteristics of the crack feature sequence.

[0107] In this embodiment, the combination of step S20 and step S30 can accurately reflect the texture features, depth features and dynamic expansion direction characteristics of bridge cracks through a comprehensive feature matrix sequence. In S20, the feature matrix sequence is constructed and screened to make the data input into the diffusion probability model targeted and relevant. This data preprocessing process effectively eliminates the interference of irrelevant features, thereby providing high-quality input data for the training and prediction of the diffusion probability model in S30, and improving the model's prediction accuracy and reliability for crack expansion rate and dynamic behavior. It can be understood that only by aligning the size and weighted fusion of the comprehensive feature matrix through mathematical transformation in S20, and screening out a highly correlated feature matrix sequence in combination with the crack direction range, can representative and complete input features be provided for the diffusion probability model in S30. This sequential step combination ensures the efficiency and accuracy of dynamic crack monitoring, and provides more accurate support for the real-time prediction of crack expansion rate and morphological changes.

[0108] This embodiment constructs a reflection amplitude matrix and a crack depth matrix based on ultrasonic data, obtains contrast values, homogeneity values ​​and correlation values ​​in the bridge crack image based on the grayscale symbiosis algorithm, constructs a texture feature matrix based on the contrast values, homogeneity values ​​and correlation values, constructs W comprehensive feature matrices based on the reflection amplitude matrix, the crack depth matrix and the texture feature matrix, determines a feature matrix sequence based on the W comprehensive feature matrices, inputs the feature matrix sequence into a pre-trained diffusion probability model, and obtains crack monitoring results output by the diffusion probability model. This embodiment combines the image texture features of ultrasonic data and the bridge crack image to respectively construct a reflection amplitude matrix, a crack depth matrix and a texture feature matrix, comprehensively characterizes the dynamic change characteristics of the bridge crack, and solves the problem of insufficient accuracy and reliability of crack feature extraction in the prior art. By screening the feature matrix sequence based on the W comprehensive feature matrices and inputting it into the pre-trained diffusion probability model, the crack expansion rate can be captured in real time, and the dynamic change of the crack can be monitored, which solves the problem of insufficient dynamic monitoring capability in the prior art.

[0109] Example 2

[0110] See also Figure 2 As shown, based on the same inventive concept, this embodiment discloses a bridge crack dynamic monitoring system based on a diffusion probability model. For details not provided in this embodiment, please refer to the description of the relevant parts in Embodiment 1. The system includes:

[0111] Data processing module: used to obtain ultrasonic data and bridge crack images, construct reflection amplitude matrix and crack depth matrix according to ultrasonic data, and obtain contrast value, homogeneity value and correlation value in bridge crack images according to grayscale symbiosis algorithm;

[0112] Methods for acquiring ultrasonic data include:

[0113] The bridge crack images are classified to obtain the corresponding crack classification results. If the crack classification result is a longitudinal crack, the ultrasonic probe is arranged along the length direction of the bridge. If the crack classification result is a transverse crack, the ultrasonic probe is arranged perpendicular to the length direction of the bridge. If the crack classification result is an oblique crack, the ultrasonic probe is arranged along the crack direction of the bridge.

[0114] The method of obtaining the contrast value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0115] The contrast value in the direction θ and the co-occurrence probability in the direction θ in the bridge crack image are obtained according to the grayscale co-occurrence algorithm, and the corresponding contrast value is calculated according to the contrast value in the direction θ and the co-occurrence probability in the direction θ.

[0116] The method of calculating the corresponding contrast value according to the contrast value in the direction θ and the co-occurrence probability in the direction θ includes:

[0117] Cot θ =∑ i,j (ij) 2 P θ (i,j);

[0118] Among them, Cot θ is the contrast value in direction θ, P θ (i, j) represents the co-occurrence probability of a pixel pair with grayscale value i and grayscale value j in direction θ.

[0119] The method of obtaining the homogeneity value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0120]

[0121] Among them, Hoy θ is the homogeneity value in the direction θ.

[0122] The method for obtaining the correlation value in the bridge crack image according to the grayscale symbiosis algorithm includes:

[0123] According to the grayscale symbiosis algorithm, the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j in the bridge crack image are obtained, and the corresponding correlation value is calculated according to the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j.

[0124] The method of calculating the corresponding correlation value according to the mean of the gray value i, the mean of the gray value j, the standard deviation of the gray value i, and the standard deviation of the gray value j includes:

[0125]

[0126] Among them, Con θ is the correlation value in direction θ, μ i Represents the mean value of gray value i, μ j Characterizes the mean of the gray value j, σ i Characterizes the standard deviation of gray value i, σ j Characterizes the standard deviation of the gray value j.

[0127] Matrix construction module: used to construct a texture feature matrix based on contrast value, homogeneity value and correlation value, construct W comprehensive feature matrices according to reflection amplitude matrix, crack depth matrix and texture feature matrix, and determine feature matrix sequence based on W comprehensive feature matrices;

[0128] Methods for constructing a texture feature matrix based on contrast values, homogeneity values, and correlation values ​​include:

[0129] The contrast values, homogeneity values ​​and correlation values ​​in K directions in the bridge crack image are calculated according to the grayscale symbiosis algorithm, and the contrast values, homogeneity values ​​and correlation values ​​in different directions are combined respectively to construct the texture feature matrix.

[0130] The method for determining the feature matrix sequence based on W comprehensive feature matrices includes:

[0131] The crack direction range in the bridge crack image is determined according to the Hough transform algorithm, and P comprehensive feature matrices are screened out from W comprehensive feature matrices according to the crack direction range, and the P comprehensive feature matrices are used as a feature matrix sequence. The crack direction range is the range of the crack extension direction in the bridge crack image.

[0132] The method of selecting P comprehensive feature matrices from W comprehensive feature matrices according to the crack direction range includes:

[0133] The value range corresponding to each comprehensive feature matrix is ​​obtained to determine whether the value range belongs to the crack direction range. If not, the corresponding comprehensive feature matrix is ​​eliminated to screen out P comprehensive feature matrices, and the value range is the value range of K.

[0134] Monitoring module: used to input the feature matrix sequence into the pre-trained diffusion probability model to obtain the crack monitoring result output by the diffusion probability model, wherein the crack monitoring result at least includes the crack propagation rate;

[0135] The training methods of the diffusion probability model include:

[0136] Pair the historical feature matrix sequence with the corresponding historical crack monitoring result label to construct a training set. The historical feature matrix sequence is used as the input of the sub-model, and the corresponding historical crack monitoring result is used as the output of the sub-model. The sub-model can be U-Net or Transformer.

[0137] The total number of diffusion steps is preset to T, and the noise increment of each step is defined as β t For each data point in the training set, the current diffusion step t and the corresponding noise data x are generated by random sampling. t , according to the current diffusion step t and the noise data x t Construct the loss function, the noise data x t The distribution form of is Gaussian distribution;

[0138] Randomly extract a batch of data points from the training set and generate corresponding noise data x for each data point t, calculate the corresponding loss function value, update the sub-model parameters through the back-propagation algorithm and minimize the loss function value as the goal, and iteratively train the sub-model to obtain the diffusion probability model.

[0139] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.

[0140] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD) or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0141] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0143] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0144] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0146] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0147] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A bridge crack dynamic monitoring method based on a diffusion probability model, characterized in that: include: Acquire ultrasonic data and bridge crack images, construct reflection amplitude matrix and crack depth matrix according to the ultrasonic data, and obtain contrast value, homogeneity value and correlation value in the bridge crack image according to the grayscale symbiosis algorithm; A texture feature matrix is ​​constructed based on contrast values, homogeneity values ​​and correlation values, W comprehensive feature matrices are constructed based on the reflection amplitude matrix, the crack depth matrix and the texture feature matrix, and a feature matrix sequence is determined based on the W comprehensive feature matrices; The characteristic matrix sequence is input into a pre-trained diffusion probability model to obtain a crack monitoring result output by the diffusion probability model, wherein the crack monitoring result at least includes a crack propagation rate.

2. The bridge crack dynamic monitoring method based on the diffusion probability model according to claim 1 is characterized in that: The method for acquiring ultrasonic data comprises: The bridge crack images are classified to obtain the corresponding crack classification results. If the crack classification result is a longitudinal crack, the ultrasonic probe is arranged along the length direction of the bridge. If the crack classification result is a transverse crack, the ultrasonic probe is arranged perpendicular to the length direction of the bridge. If the crack classification result is an oblique crack, the ultrasonic probe is arranged along the crack direction of the bridge.

3. The bridge crack dynamic monitoring method based on the diffusion probability model according to claim 2 is characterized in that: The ultrasonic data includes the peak occurrence time, and the method for constructing the crack depth matrix according to the ultrasonic data includes: The crack depth value is determined according to the peak occurrence time in the ultrasonic data and the preset propagation speed, and the crack depth matrix is ​​constructed according to the crack depth value.

4. The bridge crack dynamic monitoring method based on the diffusion probability model according to claim 1 is characterized in that: The method for obtaining the contrast value in the bridge crack image according to the grayscale symbiosis algorithm comprises: The contrast value in the direction θ and the co-occurrence probability in the direction θ in the bridge crack image are obtained according to the grayscale co-occurrence algorithm, and the corresponding contrast value is calculated according to the contrast value in the direction θ and the co-occurrence probability in the direction θ.

5. The bridge crack dynamic monitoring method based on the diffusion probability model according to claim 4 is characterized in that: The method for obtaining the correlation value in the bridge crack image according to the grayscale symbiosis algorithm includes: According to the grayscale symbiosis algorithm, the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j in the bridge crack image are obtained, and the corresponding correlation value is calculated according to the mean of the grayscale value i, the mean of the grayscale value j, the standard deviation of the grayscale value i and the standard deviation of the grayscale value j.

6. The bridge crack dynamic monitoring method based on the diffusion probability model according to claim 5 is characterized in that: The method for constructing W comprehensive feature matrices according to the reflection amplitude matrix, the crack depth matrix and the texture feature matrix comprises: Adjust the value of K, obtain W texture feature matrices, traverse the W texture feature matrices, align the reflection amplitude matrix, crack depth matrix and texture feature matrix by mathematical transformation, perform weighted fusion on the reflection amplitude matrix, crack depth matrix and texture feature matrix, and construct W comprehensive feature matrices, where K represents the value angle corresponding to the texture feature matrix.

7. The bridge crack dynamic monitoring method based on diffusion probability model according to claim 6 is characterized in that: The method for determining a feature matrix sequence based on W comprehensive feature matrices comprises: The crack direction range in the bridge crack image is determined according to the Hough transform algorithm, and P comprehensive feature matrices are screened out from W comprehensive feature matrices according to the crack direction range, and the P comprehensive feature matrices are used as a feature matrix sequence. The crack direction range is the range of the crack extension direction in the bridge crack image.

8. The bridge crack dynamic monitoring method based on diffusion probability model according to claim 7 is characterized in that: The method of selecting P comprehensive feature matrices from W comprehensive feature matrices according to the crack direction range includes: The value range corresponding to each comprehensive feature matrix is ​​obtained to determine whether the value range belongs to the crack direction range. If not, the corresponding comprehensive feature matrix is ​​eliminated to screen out P comprehensive feature matrices, and the value range is the value range of K.

9. The bridge crack dynamic monitoring method based on diffusion probability model according to claim 8 is characterized in that: The training method of the diffusion probability model includes: Pair the historical feature matrix sequence with the corresponding historical crack monitoring result label to construct a training set, use the historical feature matrix sequence as the input of the sub-model, and use the corresponding historical crack monitoring result as the output of the sub-model; The total number of diffusion steps is preset to T, and the noise increment of each step is defined as β t For each data point in the training set, the current diffusion step t and the corresponding noise data x are generated by random sampling. t , according to the current diffusion step t and the noise data x t Construct the loss function, the noise data x t The distribution form of is Gaussian distribution; Randomly extract a batch of data points from the training set and generate corresponding noise data x for each data point t , calculate the corresponding loss function value, update the sub-model parameters through the back-propagation algorithm and minimize the loss function value as the goal, and iteratively train the sub-model to obtain the diffusion probability model.

10. A bridge crack dynamic monitoring system based on a diffusion probability model, which is used to implement a bridge crack dynamic monitoring method based on a diffusion probability model as claimed in any one of claims 1 to 9, characterized in that: include: Data processing module: used to obtain ultrasonic data and bridge crack images, construct reflection amplitude matrix and crack depth matrix according to ultrasonic data, and obtain contrast value, homogeneity value and correlation value in bridge crack images according to grayscale symbiosis algorithm; Matrix construction module: used to construct a texture feature matrix based on contrast value, homogeneity value and correlation value, construct W comprehensive feature matrices according to reflection amplitude matrix, crack depth matrix and texture feature matrix, and determine feature matrix sequence based on W comprehensive feature matrices; Monitoring module: used to input the feature matrix sequence into the pre-trained diffusion probability model to obtain the crack monitoring result output by the diffusion probability model, and the crack monitoring result at least includes the crack propagation rate.

Images